A Deep Learning Method Approach for Sleep Stage Classification with EEG Spectrogram

International Journal of Environmental Research and Public Health, Journal Year: 2022, Volume and Issue: 19(10), P. 6322 - 6322

Published: May 23, 2022

The classification of sleep stages is an important process. However, this process time-consuming, subjective, and error-prone. Many automated methods use electroencephalogram (EEG) signals for classification. These do not classify well enough perform poorly in the N1 due to unbalanced data. In paper, we propose a stage method using EEG spectrogram. We have designed deep learning model called EEGSNet based on multi-layer convolutional neural networks (CNNs) extract time frequency features from spectrogram, two-layer bi-directional long short-term memory (Bi-LSTMs) learn transition rules between adjacent epochs stages. addition, improve generalization ability model, used Gaussian error linear units (GELUs) as activation function CNN. proposed was evaluated by four public databases, Sleep-EDFX-8, Sleep-EDFX-20, Sleep-EDFX-78, SHHS. accuracy 94.17%, 86.82%, 83.02% 85.12%, respectively, datasets, MF1 87.78%, 81.57%, 77.26% 78.54%, Kappa 0.91, 0.82, 0.77 0.79, respectively. our achieved better results N1, with F1-score 70.16%, 52.41%, 50.03% 47.26% datasets.

Language: Английский

---

A Robust Deep Learning Framework Based on Spectrograms for Heart Sound Classification

IEEE/ACM Transactions on Computational Biology and Bioinformatics, Journal Year: 2023, Volume and Issue: 21(4), P. 936 - 947

Published: Feb. 22, 2023

Heart sound analysis plays an important role in early detecting heart disease. However, manual detection requires doctors with extensive clinical experience, which increases uncertainty for the task, especially medically underdeveloped areas. This paper proposes a robust neural network structure improved attention module automatic classification of wave. In preprocessing stage, noise removal Butterworth bandpass filter is first adopted, and then recordings are converted into time-frequency spectrum by short-time Fourier transform (STFT). The model driven STFT spectrum. It automatically extracts features through four down sample blocks different filters. Subsequently, based on Squeeze-and-Excitation coordinate developed feature fusion. Finally, will give category waves learned features. global average pooling layer adopted reducing model's weight avoiding overfitting, while focal loss further introduced as function to minimize data imbalance problem. Validation experiments have been conducted two publicly available datasets, results well demonstrate effectiveness advantages our method.

Language: Английский

---

MaskSleepNet: A Cross-Modality Adaptation Neural Network for Heterogeneous Signals Processing in Sleep Staging

IEEE Journal of Biomedical and Health Informatics, Journal Year: 2023, Volume and Issue: 27(5), P. 2353 - 2364

Published: March 7, 2023

Deep learning methods have become an important tool for automatic sleep staging in recent years. However, most of the existing deep learning-based approaches are sharply constrained by input modalities, where any insertion, substitution, and deletion modalities would directly lead to unusable model or a deterioration performance. To solve modality heterogeneity problems, novel network architecture named MaskSleepNet is proposed. It consists masking module, multi-scale convolutional neural (MSCNN), squeezing excitation (SE) block, multi-headed attention (MHA) module. The module adaptation paradigm that can cooperate with discrepancy. MSCNN extracts features from multiple scales specially designs size feature concatenation layer prevent invalid redundant zero-setting channels. SE block further optimizes weights optimize efficiency. MHA outputs prediction results temporal information between sleeping features. performance proposed was validated on two publicly available datasets, Sleep-EDF Expanded (Sleep-EDFX) Montreal Archive Sleep Studies (MASS), clinical dataset, Huashan Hospital Fudan University (HSFU). achieve favorable discrepancy, e.g. single-channel EEG signal, it reach 83.8%, 83.4%, 80.5%, two-channel EEG+EOG signals 85.0%, 84.9%, 81.9% three-channel EEG+EOG+EMG signals, 85.7%, 87.5%, 81.1% Sleep-EDFX, MASS, HSFU, respectively. In contrast accuracy state-of-the-art approach which fluctuated widely 69.0% 89.4%. experimental exhibit maintain superior robustness handling discrepancy issues.

Language: Английский

---

SleepEGAN: A GAN-enhanced ensemble deep learning model for imbalanced classification of sleep stages

Biomedical Signal Processing and Control, Journal Year: 2024, Volume and Issue: 92, P. 106020 - 106020

Published: Feb. 8, 2024

Language: Английский

---

ViT-SENet-Tom: machine learning-based novel hybrid squeeze–excitation network and vision transformer framework for tomato fruits classification

Neural Computing and Applications, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 20, 2025

Language: Английский

---

Current status and prospects of automatic sleep stages scoring: Review

Biomedical Engineering Letters, Journal Year: 2023, Volume and Issue: 13(3), P. 247 - 272

Published: July 10, 2023

Abstract The scoring of sleep stages is one the essential tasks in analysis. Since a manual procedure requires considerable human and financial resources, incorporates some subjectivity, an automated approach could result several advantages. There have been many developments this area, order to provide comprehensive overview, it review relevant recent works summarise characteristics approaches, which main aim article. To achieve it, we examined articles published between 2018 2022 that dealt with stages. In final selection for in-depth analysis, 125 were included after reviewing total 515 publications. results revealed automatic demonstrates good quality (with Cohen's kappa up over 0.80 accuracy 90%) analysing EEG/EEG + EOG EMG signals. At same time, should be noted there has no breakthrough using these signals years. Systems involving other potentially acquired more conveniently user (e.g. respiratory, cardiac or movement signals) remain challenging implementation high level reliability but innovation capability. general, stage excellent potential assist medical professionals while providing objective assessment.

Language: Английский

---

SleePyCo: Automatic sleep scoring with feature pyramid and contrastive learning

Expert Systems with Applications, Journal Year: 2023, Volume and Issue: 240, P. 122551 - 122551

Published: Nov. 18, 2023

Automatic sleep scoring is essential for the diagnosis and treatment of disorders enables longitudinal tracking in home environments. Conventionally, learning-based automatic on single-channel electroencephalogram (EEG) actively studied because obtaining multi-channel signals during difficult. However, learning representation from raw EEG challenging owing to following issues: (1) sleep-related patterns occur different temporal frequency scales 2) stages share similar patterns. To address these issues, we propose an Sleep framework that incorporates a feature Pyramid supervised Contrastive learning, named SleePyCo. For pyramid, backbone network SleePyCo-backbone consider multiple sequences scales. Supervised contrastive allows extract class discriminative features by minimizing distance between intra-class simultaneously maximizing inter-class features. Comparative analyses four public datasets demonstrate SleePyCo consistently outperforms existing frameworks based EEG. Extensive ablation experiments show exhibited enhanced overall performance, with significant improvements discrimination stages, especially N1 rapid eye movement (REM). Source code available at https://github.com/gist-ailab/SleePyCo.

Language: Английский

---

Multi-dimensional stereo face reconstruction for psychological assistant diagnosis in medical meta-universe

Information Sciences, Journal Year: 2023, Volume and Issue: 654, P. 119831 - 119831

Published: Nov. 4, 2023

Language: Английский

---

DCNet: A Self-Supervised EEG Classification Framework for Improving Cognitive Computing-Enabled Smart Healthcare

IEEE Journal of Biomedical and Health Informatics, Journal Year: 2024, Volume and Issue: 28(8), P. 4494 - 4502

Published: Jan. 23, 2024

Cognitive computing endeavors to construct models that emulate brain functions, which can be explored through electroencephalography (EEG). Developing precise and robust EEG classification is crucial for advancing cognitive computing. Despite the high accuracy of supervised models, they are constrained by labor-intensive annotations poor generalization. Self-supervised address these issues but encounter difficulties in matching learning. Three challenges persist: 1) capturing temporal dependencies EEG; 2) adapting loss functions describe feature similarities self-supervised models; 3) addressing prevalent issue data imbalance EEG. This study introduces DreamCatcher Network (DCNet), a framework with two-stage training strategy. The first stage extracts representations contrastive learning, second transfers representation encoder task. DCNet utilizes time-series learning autonomously comprehensively capture correlations. A novel function, SelfDreamCatcherLoss, proposed evaluate between enhance performance DCNet. Additionally, two augmentation methods integrated alleviate class imbalances. Extensive experiments show superiority over current state-of-the-art achieving on both Sleep-EDF HAR datasets. It holds substantial promise revolutionizing sleep disorder detection expediting development advanced healthcare systems driven

Language: Английский

---

Recognition of Converter Steelmaking State Based on Convolutional Recurrent Neural Networks

Metallurgical and Materials Transactions B, Journal Year: 2024, Volume and Issue: 55(3), P. 1856 - 1868

Published: March 28, 2024

Language: Английский

---